Radionuclide Transport and Uptake in Coastal Aquatic Ecosystems: A Comparison of a 3D Dynamic Model and a Compartment Model

The coastal area considered in this study is a sub-basin of a shallow coastal bay in the Baltic Proper near Forsmark, Sweden. The area has a surface area of 11.5 km² and is known as “basin 116,” one of 28 basins (“biosphere objects”) in the Forsmark area (Brydsten 2006; Lindborg et al. 2013) used in the safety assessment of a geological repository for spent nuclear fuel. Its location is shown in Kautsky et al. (2013, their Fig. 3). The photo in Fig. 1 provides a general view of the area. The marine ecosystems at the site and other important site data are described in Aquilonius (2010), who summarizes site data such as hydrodynamics, chemical and physical characteristics, biota types and biomass, as well as quantification of ecosystem processes. Elemental transfers in this area have also been studied (Bradshaw et al. 2012).

The K-model is an ecological marine radionuclide transport model of a coastal food web developed by Kumblad et al. (2003, 2006), explained detail in Kumblad et al. (2003) and Kumblad and Kautsky (2004). For this study, the K-model has been improved and applied in the newer software package Ecolego 5 (Broed and Xu 2008). It is a food-web model that describes the biomass distribution and the carbon dynamics of the ecosystem, and includes both biotic and abiotic compartments. Radionuclides follow the flow of carbon in the ecosystem.

The 3D hydrodynamic MIKE3-FM model (Graham and Butts 2005; Butts and Graham 2008) and associated ecosystem model developed in ECOLab (D-model) were adapted and set-up to represent the specific water exchange and ecosystem conditions in Forsmark area. Resolution of the model varies from less than 20 m (near the coast) to more than 100 m in the open part of the model domain. A detailed description of the hydrodynamic model is found in Karlsson et al. (2010) and Eriksson and Engqvist (2013). The ecosystem and radionuclide food-web models for the present-day situation were implemented in MIKE using the ECO Lab software (DHI 2011), based mainly on in situ data collected during a single year (2004). Conceptually, the ecosystem model and the radionuclide transport model have been developed based on the general food-web structure introduced in earlier modeling studies within the area (e.g., Kumblad and Kautsky 2004), and for this article the high-resolution analysis considers only a limited number of selected radionuclides (Box S1, Electronic Supplementary Material).

In the models, radionuclides are released into the environment from a point source (K-model) or number of distributed sources (D-model). The release of radionuclides is set at a constant rate of 1 Bq of each radionuclide per year. In the D-model, the sources were distributed according to the results of the groundwater modeling described in Berglund et al. (2013) and are mainly situated in basins 116, 117, 118, 120, and 121 (locations are shown in Fig. 3 of Kautsky et al. 2013). The sum of all the sources equals 1 Bq of each radionuclide per year. In the K-model, instantaneous homogeneous mixing of the radionuclides in the entire water volume is assumed. Hence, the point source in the K-model is included in the water phase directly, whereas the various sources in the D-model release the radionuclide to the sediment pore water from where it moves through the sediment to the water phase. The process of radioactive decay is not included in this study because of its minor importance for the environmental fate of these long-lived radionuclides. However, in the assessment model, decay and important decay chains are handled (Avila et al. 2013).

The two models share several features, such as the identity of compartments and state variables of the ecological model structure (Table 1), but the models differ greatly in how the underlying ecological models are executed. The structure and rates of the K-model are built on site-specific measurements of the biomasses of various functional groups and key species (see Kumblad et al. 2003, 2006) and corresponding rates of ecological processes, whereas the D-model is an ecosystem model (DHI 2011) that is adaptable to any aquatic ecosystem and has been used in numerous studies, and, for this study, was supplemented with four additional functional groups: planktivorous fish, deposit feeders, herbivores, and benthic predators.

Figure 2 shows schematic (simplified) structures of the two ecosystem models. The K-model operates with space-averaged biomasses and rates taking into account depth-dependent variations in the photic zone, and uses parameters (e.g., insolation, temperature) integrated over 1 year. Hence, the K-model maintains constant biomasses through simulation years. The K-model groups species having the same ecological functions into one biological compartment, thus reducing the number of state variables. The K-model is not calibrated, but the primary production is adjusted to a separately run-coupled nutrient model (Kumblad and Kautsky 2004) that takes into account the nutrients and water exchange across the boundaries (Eriksson and Engqvist 2013).

The two models share most of the same state variables or groups of state variables (Table 1), with the following exceptions: In the D-model, the effect of benthic filter-feeders was imposed by a spatially varying filtration of phytoplankton and particulate organic matter in the near-bed model layer according to the measured distribution of mussels, cockles, and clams. The effect of heterotrophic bacteria and meiofauna are implemented in the D-model as mineralization of carbon and nutrients, driven by the concentration and supply of particulate organic matter (POM) and temperature. Overall, the D-model is driven by light availability (insolation, light attenuation—including the effect of resuspension), nutrient availability (run-off, atmospheric deposition, sediment, and water column mineralization), and exchange across model boundaries (wind- and water-level-driven circulation). The D-model was calibrated against measurements of nutrients, chlorophyll, Secchi depth, and biomasses of benthic vegetation and deposit feeders (Erichsen et al. 2010).

In both models, radionuclides are assumed to follow the flow of organic carbon in the food web, and radionuclide relocation is regulated by several radionuclide-specific mechanisms: uptake by phytoplankton and benthic vegetation, adsorption to organic surfaces, and assimilation and excretion by animals. In contrast to the K-model, the D-model considers adsorption of radionuclides to suspended matter and sediments with different partition coefficients, K _ds, for organic and inorganic particles. The D-model includes a sediment module consisting of two compartments with an upper active layer subject to re-suspension and a lower layer consisting of consolidated sediments. In sediments, radionuclides are adsorbed to organic and inorganic matters, dissolved in pore water and, depending on (modeled) oxygen penetration and redox conditions, certain radionuclides may precipitate (as sulfides or carbonates) or dissolve.

In both models, radionuclide-specific dynamics depend on partition coefficients, K _ds, between radionuclides adsorbed to suspended particulate matter or organisms and in the surrounding water (see Box S1, Electronic Supplementary Material). As adsorption is directly proportional to the surface area, organisms with a high surface-to-volume ratio (S:V), such as phytoplankton, show high partition coefficients for adsorbing radionuclides. The S:V ratio for the different trophic state variables was calculated from the physical dimensions of dominant species within each trophic group assuming spherical, cylindrical, or flat geometrical forms according to their morphology. For fish, the total area of gills was additionally used to represent the adsorbing area (Pauly 1981). Total surface area within trophic groups was calculated from S:V ratio, biomass, and abundance based on monitoring data from Aquilonius (2010).

In both the K- and D-models, accumulated radionuclides are retained, and release takes place only when organisms die or are consumed. In the K-model, radionuclides have an additional efflux from organisms along with excreted feces.

Similarities and differences between the models are evaluated in this article by comparing modeled CRs, defined as the ratio of the radionuclide concentration in the organism resulting from all exposure pathways (including water, sediment, and dietary pathways) to the concentration in sea water (or pore water for infauna), normalized to the carbon content of the biota fraction and expressed as [(Bq kg⁻¹ fw)/(kg_C per kg fw)] per (Bq m⁻³), i.e., kg_C m⁻³, where fw is fresh weight. The CR definition assumes that the radionuclides in the organisms are in equilibrium with the ambient sea water. A large number of radionuclides that could hypothetically be released from the repository of radioactive waste have been modeled (Kumblad et al. 2006; Avila et al. 2010, 2013; Erichsen et al. 2010), but for this article, we have selected three radionuclides for comparison of model results: ¹³⁵Cs, ⁵⁹Ni, and ²³⁰Th. All three are long-lived radionuclides relevant to the planned high-level radioactive waste repository. Cs is also of interest due to its post-Chernobyl abundance as ¹³⁷Cs in the Baltic Sea (HELCOM 2009). The element Ni is also biologically interesting as it is essential for many phytoplankton species (Muyssen et al. 2004). ²³⁰Th can be a naturally occurring daughter of ²³⁸U decay which is highly particle reactive, i.e., high K _d, and does not assimilate in organic tissues. In contrast, Cs and Ni have around ten times lower K _d values than Th.

The D-model showed gradients in radionuclide concentrations in water and of CR in phytoplankton (Fig. 3), from the repository outlet to the open boundary of the Forsmark area. For example, within the Forsmark area the yearly averaged CR for Cs-135 in phytoplankton varied by a factor of 10–20, and this variation is explained primarily by the spatial variation in dissolved ¹³⁵Cs concentration (Fig. 3). Even at smaller spatial scales (e.g., within basin 116) the D-model shows a threefold variation in yearly averaged CR for phytoplankton (Fig. 3).

In addition to spatial variation, the radionuclide concentration and CR in phytoplankton vary on a weekly timescale driven by water exchange, resulting in alternating conditions with water and plankton containing high concentrations (i.e., associated with the radionuclide plume) followed by low concentrations when uncontaminated water and plankton are passing the fixed site. Whereas the modeled phytoplankton biomass has a distinct seasonal variation (Fig. 4A), the seasonal variation in CR is somehow smaller with about two times higher CRs during winter than during summer (Fig. 4C), suggesting a negative influence of phytoplankton biomass on CR. Some of the variation in CR is explained by a varying turnover of phytoplankton biomass and the “slow” adsorption–desorption processes resulting in a loose coupling between changes in radionuclide concentrations in water and phytoplankton. Partial decoupling of ¹³⁵Cs concentration in water and phytoplankton is especially visible during the spring bloom in April (Fig. 4B). The same patterns as shown in Fig. 4, though with different CRs, are found for the other radionuclides and other sites within the study area. Mass balance modeling and empirical measurements in the area also show this imbalance for many elements during the spring bloom (Bradshaw et al. 2012).

It is thus clear that the CR approach has major pitfalls when working with releases from point sources: (i) organisms are diluted with unpolluted ones and (ii) spatial and temporal distributions are not homogeneous in the case of a point source. This is not usually taken into account in risk assessment because measured CRs are for elements which are homogeneously distributed in the environment, i.e., there is no dilution by uncontaminated organisms, or they are measured in the laboratory, disregarding all these processes.

The results of the two model simulations of the CR of phytoplankton, zooplankton, and fish for the three elements ¹³⁵Cs, ⁵⁹Ni, and ²³⁰Th are presented in Table 2 and in Fig. 5, where they are compared with site-specific CR values. Overall, the predicted results from models K and D were in reasonable agreement considering the differences in applied distribution constants and model structures.

Despite differences in the model structures and input constants, modeled CR values for the three organism groups and three elements presented in this article were in reasonable agreement, suggesting that both models are robust. However, the two different models have strengths and weaknesses regarding realism and practicability that make them more or less applicable to different needs. The D-model allows estimates at a high spatial and temporal resolution (down to 20 m and over a few hours). This gives insights in how potential hotspots of exposure in space and time (e.g., algal blooms and localized radionuclide releases) can be identified or modeled, and thus a better assessment made of risks to organisms in the ecosystem and nearby areas.

However, the complex numerical calculations cause long computational times and thus limit the simulation period to decades, not centuries. Such precision may not always be necessary in a risk assessment, but is useful when specific events, measurements, or local heterogeneity need to be more accurately considered and assessed on shorter time scales (Harms et al. 2003). The high resolution gives an estimate of how large the errors are with coarser resolutions and helps to explore how, e.g., CR varies over space and time. The D-model is constrained by the mass balance of nutrients and organic matter, which means that estimated ecosystems are scaled properly to carbon- and water turnover.

Compartment models such as the K-model do not include hydrodynamics explicitly, but are based on the assumption that average water retention time from hydrodynamic models and/or field data are sufficiently good approximations for estimating the annual dynamics of the ecosystem. The K-model also assumes instantaneous and homogeneous distribution (over a year) of the radionuclides released within the compartment. This is not the case in reality, but seems to be of minor importance, varying the CR within the compartment within a factor of three. Variation of parameters in space and time is taken into account by consideration of their probability distribution functions, and applying the probabilistic simulations in the box model is much quicker and easier than in the 3D model. This also enables the K-model to make estimates over a longer time period (i.e., >100 years), which is necessary for the risk assessment of long-lived radionuclides. The structure of the K-model is such that it can be run at any spatial or temporal scale comparable to that of the D-model.

However, it requires input parameters for water exchange scaled to the same area that requires a water exchange model which is similar to MIKE-3, as used for the D-model. The mass balance of organic matter in the current K-model needs to be scaled by nutrient mass balance from a separate model (Kumblad and Kautsky 2004). Thus, the K-model requires more effort to manually scale the input parameters to proper temporal and spatial contexts.

A strength of both models is that they allow the incorporation of many biological compartments, food-web processes, and specific interactions and transfers such as feeding, respiration, death, excretion, etc., adding to the ecological realism of the model. In addition, they allow the calculation of the range and confidence limits of output values such as CR values, instead of simply producing mean values. This both allows the degree of confidence in the values to be assessed and also enables the identification of potentially high values that may lead to unwanted exposures.

The two models have been developed separately, but could be used to complement each other in different ways:

The mechanistic approach used in the models also shows that the need for element-specific data for many organisms can be reduced to favor more site-specific, high quality, and well-controlled measurements of a few organisms. The D-model also shows that the traditional approach with CR models has several flaws regarding the nature of point sources, and shows that CR measurements should be treated carefully with respect to temporal and spatial representations. Both models are based on substantial amounts of field data from the Forsmark site. The D-model uses flow fields based on real topography of the Forsmark area and realistic forcing functions such as wind and density, and both models use field data on organism biomass distributions, insolation, etc. This undoubtedly contributes to their robustness, but also requires substantial underlying scientific and financial input. This is not always possible in all risk assessment contexts, but in large-scale investigations such as those required prior to the building of a deep geological repository, much of this data must be collected. Thus, by proper planning and project management any extra data required for such modeling can be obtained at marginal extra effort and cost.